Fix fresh-install failures in predictive/GNN templates by pinning to a version with the gnn extra

[somacdivad]

Source and scope

A user hit ModuleNotFoundError: No module named 'relationalai_gnns' while doing a fresh install of the smoker_status_prediction template and flagged that the predictive/GNN templates are pinned to relationalai versions that predate the gnn dependency bundle.

• Slack: Heads up: the predictive/GNN templates are pinned to incompatible relationalai versions

The relationalai[gnn] extra (which pulls in relationalai-gnns) was introduced in relationalai 1.16.0. Every predictive template was still pinned below that (1.8, 1.11.0, or 1.15.0), so a clean install never installs the GNN runtime and fails at the first GNN import.

This PR partially addresses the thread — it fixes the version-pin bug only. The other topics raised in the thread (removing the historical v0.13/v0.14 snapshot folders, flattening v1 into the repo root, and merging the v1 template index into the root README) are intentionally left for follow-up work.

What changed

Pinned all nine predictive/GNN v1 templates to relationalai[gnn]==1.27.1 — the latest release that provides the gnn extra — keeping the repo's exact-==-pin convention.

• pyproject.toml (dependencies): updated in retail_planning, memory_supply_allocation, energy_grid_planning, subscriber_retention, telco_network_recovery, datacenter_compute_allocation, fraud-detection, smoker_status_prediction, demand_forecasting.
• README.md prerequisite/reproducibility lines: updated to match the new pin in the same templates (where a version was stated).

Why

Before: the templates declared relationalai[gnn]==1.8|1.11.0|1.15.0. Those versions have no gnn extra, so a fresh pip install of the template silently omits relationalai-gnns and the run dies with ModuleNotFoundError: No module named 'relationalai_gnns'.

After: the templates declare a version that actually ships the gnn extra, so a clean install pulls in the GNN runtime and the predictive step can run.

Risks

• The templates were originally authored against older SDKs (1.8–1.15). Moving to 1.27.1 is a significant jump; if any predictive template relies on GNN/predictive API behavior that changed between those releases, that template's script may need follow-up adjustment. The template scripts were not executed end-to-end as part of this change.

How to verify

For any predictive template (e.g. v1/smoker_status_prediction):

1. Create a clean virtualenv and pip install . from the template directory.
2. Confirm relationalai-gnns is installed (pip show relationalai-gnns).
3. Run the template's *.py entrypoint and confirm it no longer fails with ModuleNotFoundError: No module named 'relationalai_gnns'.
4. Spot-check that no predictive template still references a pre-1.16 pin: grep -rn "relationalai\[gnn\]" v1 should show only 1.27.1.

[github-actions]

The templates docs preview for this pull request has been deployed to Vercel!

✅ Preview:	https://relationalai-docs-qv0ug9p39-relationalai.vercel.app/build/templates
🔍 Inspect:	https://vercel.com/relationalai/relationalai-docs/3y9E789EEEkXBXkqnLooUUL3E2Wt